A power efficient crossbar arbitration in multi-NoC for multicast and broadcast traffic

Large scale chip multiprocessors employ a multi-NoC, consisting of multiple physical channels for inter-core communication. Placement of a custom arbitration logic can improve the critical path delay and relax the worst case timing closure of the network. In particular, it can effectively distribute and manage the traffic from the multi-threaded workloads among the multiple networks of the NoC. This paper gives the design and implementation of the arbitration logic at the router crossbars. The results are compared with baseline NoC and other multi-NoC architectures. The proposed energy efficient router saves up to 57% of the router power consumption.